I'm using Fisher-faces and I would like to combined with Eigenvalues face recognition to have a more accurate prediction how would i do this? I keep getting false predictions. When I train my face recognition what is best in pictures sizes of images and file types? I would to be able to recognize my class mates when they walk into a room so the computer greets them as they walk in.
thanks
"I would like to combined with Eigenvalues face recognition"
that's already the case. fisherfaces are an extension of eigenfaces ( additional LDA )
"what is best in pictures sizes of images and file types ?"
stick with .png or .pgm, avoid jpg. resize to like 100x100.
"I keep getting false predictions"
make sure, that you're using exactly the same preprocessing for your train and test images. if your test-images are cropped from a cascade face detection, your train images should be, too. applying equalizeHist() should give some gain, too.
you'll need like 10-30 images per person. allow some variation in pose / lighting
try the LBPH face recognizer, too.
try to control the lighting in the room. if there's a lamp, that ppl can switch on or off; - one half of the pics will be with lamp, and the other without, - remove it ;)
I guess i don't know how to fix my code according to what you said because its not working can you take a look at my code on github at this url
i'm detecting faces with detectmultiscale with lbpcascade_frontalface.xml and lbp_cascade.detectMultiScale(gray, faces2, 3, 1, 0|CASCADE_DO_CANNY_PRUNING,Size(30,30)); then looping and making the prediction but i get the wrong prediction. i have applied the equalizeHist as mentioned above. my images are 112 x 92 same as the at&t database but using png as you mentioned above.
Asked: 2013-10-14 19:23:38 -0600
Seen: 840 times
Last updated: Oct 15 '13
